In [1]:
%matplotlib inline
import pandas as pd
import numpy as np
import plotly.express as px
import matplotlib.pyplot as plt
In [2]:
#url = 'https://twelvedata.com/markets/256458/crypto/huobi/eth-btc/historical-data?start_date=2025-09-01&end_date=2025-09-03&interval=30min'  # Replace with your URL
#tables = pd.read_html(url)
#url1 = 'https://twelvedata.com/markets/256458/crypto/huobi/eth-btc/historical-data?start_date=2025-09-01&end_date=2025-09-03&interval=30min&page=2'
#tables1 = pd.read_html(url1)
##tbls = pd.concat([tables[0], tables1[0]])
#tbls.to_csv('huobi_eth_bts.csv')
In [3]:
#from google.colab import drive
#drive.mount('/content/gdrive', force_remount=True)
In [4]:
#ls gdrive/MyDrive/
#%cd /content/gdrive/MyDrive/Colab Notebooks
In [5]:
##huobi trades data for auxiliary graph
tbls = pd.read_csv("huobi_eth_bts.csv")
In [6]:
tbls['timestamp'] = pd.to_datetime(tbls['Date'].values, utc=True)
In [7]:
tbls.head()
Out[7]:
Date Open High Low Close % Change timestamp
0 Sep 03, 2025 23:30 0.039891 0.039893 0.039827 0.039827 -0.16% 2025-09-03 23:30:00+00:00
1 Sep 03, 2025 23:00 0.039893 0.039932 0.039891 0.039891 -0.01% 2025-09-03 23:00:00+00:00
2 Sep 03, 2025 22:30 0.039845 0.039893 0.039845 0.039893 0.12% 2025-09-03 22:30:00+00:00
3 Sep 03, 2025 22:00 0.039820 0.039968 0.039820 0.039845 0.06% 2025-09-03 22:00:00+00:00
4 Sep 03, 2025 21:30 0.039901 0.039901 0.039753 0.039820 -0.20% 2025-09-03 21:30:00+00:00
In [8]:
#%cd /home
In [9]:
#%pwd

Preliminaries

In [10]:
#from google.colab import files

#uploaded = files.upload()
In [11]:
#for k in uploaded:
#  print(k)
In [12]:
#import io
#trds = pd.read_csv(io.StringIO(uploaded['eth-btc-trades.csv'].decode('utf-8')))
#lob    = pd.read_csv(io.StringIO(uploaded['eth-btc-orderbooks.csv'].decode('utf-8')))
In [13]:
trds = pd.read_csv('eth-btc-trades.csv')
lob    = pd.read_csv('eth-btc-orderbooks.csv')
In [14]:
trds.shape, lob.shape
Out[14]:
((845, 4), (188, 3))
In [15]:
lob.head()
Out[15]:
timestamp asks bids
0 2025-09-01 00:13:09.511132300+00:00 [{'price': 0.04100829, 'size': 0.02288708}, {'... [{'price': 0.0405, 'size': 0.00249086}, {'pric...
1 2025-09-01 00:31:23.134173200+00:00 [{'price': 0.04097411, 'size': 0.02047798}, {'... [{'price': 0.0405, 'size': 0.00249086}, {'pric...
2 2025-09-01 00:49:37.173137500+00:00 [{'price': 0.04091134, 'size': 0.02229217}, {'... [{'price': 0.0405, 'size': 0.00249086}, {'pric...
3 2025-09-01 01:07:51.274078600+00:00 [{'price': 0.04103079, 'size': 0.07205901}, {'... [{'price': 0.0405, 'size': 0.00249086}, {'pric...
4 2025-09-01 01:26:07.248503+00:00 [{'price': 0.04100837, 'size': 0.02019721}, {'... [{'price': 0.0405, 'size': 0.00249086}, {'pric...
In [16]:
np.unique([len(eval(i)) for i in lob.bids]), np.unique([len(eval(i)) for i in lob.asks])
Out[16]:
(array([50]), array([50]))
In [17]:
fun = lambda x: {'ask_'+k+'_'+str(ind): v for ind, d in enumerate(eval(x)) for k, v in d.items()}
df_asks = pd.DataFrame.from_records(lob.asks.apply(fun))
In [18]:
df_asks.head()
Out[18]:
ask_price_0 ask_price_1 ask_price_10 ask_price_11 ask_price_12 ask_price_13 ask_price_14 ask_price_15 ask_price_16 ask_price_17 ... ask_size_45 ask_size_46 ask_size_47 ask_size_48 ask_size_49 ask_size_5 ask_size_6 ask_size_7 ask_size_8 ask_size_9
0 0.041008 0.04109 0.04152 0.04157 0.0416 0.04162 0.04167 0.041696 0.0417 0.04172 ... 0.002672 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286
1 0.040974 0.04109 0.04152 0.04157 0.0416 0.04162 0.04167 0.041696 0.0417 0.04172 ... 0.002672 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286
2 0.040911 0.04109 0.04152 0.04157 0.0416 0.04162 0.04167 0.041696 0.0417 0.04172 ... 0.002672 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286
3 0.041031 0.04109 0.04152 0.04157 0.0416 0.04162 0.04167 0.041696 0.0417 0.04172 ... 0.002672 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286
4 0.041008 0.04109 0.04152 0.04157 0.0416 0.04162 0.04167 0.041696 0.0417 0.04172 ... 0.002672 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286

5 rows × 100 columns

In [19]:
fun = lambda x: {'bids_'+k+'_'+str(ind): v for ind, d in enumerate(eval(x)) for k, v in d.items()}
df_bids = pd.DataFrame.from_records(lob.bids.apply(fun))
In [20]:
df_bids.head()
Out[20]:
bids_price_0 bids_price_1 bids_price_10 bids_price_11 bids_price_12 bids_price_13 bids_price_14 bids_price_15 bids_price_16 bids_price_17 ... bids_size_45 bids_size_46 bids_size_47 bids_size_48 bids_size_49 bids_size_5 bids_size_6 bids_size_7 bids_size_8 bids_size_9
0 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.003381 0.014934 0.003115 0.021893 0.004384 0.003156 0.002522 0.013491 0.002794 0.000311
1 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.003381 0.014934 0.003115 0.021893 0.004384 0.003156 0.002522 0.013491 0.002794 0.000311
2 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.003381 0.014934 0.003115 0.021893 0.004384 0.003156 0.002522 0.013491 0.002794 0.000311
3 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.003381 0.014934 0.003115 0.021893 0.004384 0.003156 0.002522 0.013491 0.002794 0.000311
4 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.003381 0.014934 0.003115 0.021893 0.004384 0.003156 0.002522 0.013491 0.002794 0.000311

5 rows × 100 columns

In [21]:
df_bids_asks = pd.concat([df_bids, df_asks], axis=1)
df_bids_asks['timestamp'] = pd.to_datetime(lob.timestamp.values)
df_bids_asks.head()
Out[21]:
bids_price_0 bids_price_1 bids_price_10 bids_price_11 bids_price_12 bids_price_13 bids_price_14 bids_price_15 bids_price_16 bids_price_17 ... ask_size_46 ask_size_47 ask_size_48 ask_size_49 ask_size_5 ask_size_6 ask_size_7 ask_size_8 ask_size_9 timestamp
0 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286 2025-09-01 00:13:09.511132300+00:00
1 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286 2025-09-01 00:31:23.134173200+00:00
2 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286 2025-09-01 00:49:37.173137500+00:00
3 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286 2025-09-01 01:07:51.274078600+00:00
4 0.0405 0.04049 0.039919 0.0399 0.039809 0.0398 0.0397 0.0395 0.0394 0.039398 ... 0.000237 0.000237 0.002891 0.000237 0.002478 0.005148 0.029822 0.002472 0.056286 2025-09-01 01:26:07.248503+00:00

5 rows × 201 columns

In [22]:
df_list = []
for i in range(50):
  pr_name = 'bids_price_'+str(i)
  sz_name = 'bids_size_' +str(i)
  df1 = df_bids_asks[['timestamp', pr_name, sz_name]].melt(id_vars=['timestamp', sz_name], value_vars=pr_name)
  df_list.append(df1.rename(columns={sz_name: 'size'}))

df_melt_bids = pd.concat(df_list)
In [23]:
df_list = []
for i in range(50):
  pr_name = 'ask_price_'+str(i)
  sz_name = 'ask_size_' +str(i)
  df1 = df_bids_asks[['timestamp', pr_name, sz_name]].melt(id_vars=['timestamp', sz_name], value_vars=pr_name)
  df_list.append(df1.rename(columns={sz_name: 'size'}))

df_melt_asks = pd.concat(df_list)
In [24]:
df_melt_asks.head()
Out[24]:
timestamp size variable value
0 2025-09-01 00:13:09.511132300+00:00 0.022887 ask_price_0 0.041008
1 2025-09-01 00:31:23.134173200+00:00 0.020478 ask_price_0 0.040974
2 2025-09-01 00:49:37.173137500+00:00 0.022292 ask_price_0 0.040911
3 2025-09-01 01:07:51.274078600+00:00 0.072059 ask_price_0 0.041031
4 2025-09-01 01:26:07.248503+00:00 0.020197 ask_price_0 0.041008
In [25]:
df_melt_bids_asks = pd.concat([df_melt_bids, df_melt_asks])
In [26]:
df_melt_bids_asks.shape
Out[26]:
(18800, 4)
In [27]:
df_melt_bids_asks.sort_values(by='timestamp', inplace=True)
In [28]:
trds_m = trds.copy()
trds_m['timestamp'] = pd.to_datetime(trds_m.timestamp.values)
trds_m.rename(columns={'price':'value', 'side':'variable'}, inplace=True)
#trds1['size'] = np.log(trds1['size'])
trds_m.head()
Out[28]:
timestamp value size variable
0 2025-09-01 00:02:57+00:00 0.040800 0.000441 SELL
1 2025-09-01 00:02:57+00:00 0.040700 0.002478 SELL
2 2025-09-01 00:02:57+00:00 0.040700 0.020152 SELL
3 2025-09-01 00:02:57+00:00 0.040600 0.002460 SELL
4 2025-09-01 00:03:14+00:00 0.040812 91.455503 BUY
In [29]:
df_trds_bids_asks = pd.concat([trds_m, df_melt_bids_asks])

/Users/valdemar/anaconda3/lib/python3.5/site-packages/ipykernel_launcher.py:1: FutureWarning:

Sorting because non-concatenation axis is not aligned. A future version
of pandas will change to not sort by default.

To accept the future behavior, pass 'sort=False'.

To retain the current behavior and silence the warning, pass 'sort=True'.
In [30]:
df_trds_bids_asks.sort_values(by='timestamp', inplace=True)
In [31]:
df_trds_bids_asks['size_log'] = np.log2(df_trds_bids_asks['size'].values)  -np.log2(np.min(df_trds_bids_asks['size'])) +0.01
In [32]:
df_trds_bids_asks.head()
Out[32]:
size timestamp value variable size_log
0 0.000441 2025-09-01 00:02:57+00:00 0.040800 SELL 9.631725
1 0.002478 2025-09-01 00:02:57+00:00 0.040700 SELL 12.121677
2 0.020152 2025-09-01 00:02:57+00:00 0.040700 SELL 15.145104
3 0.002460 2025-09-01 00:02:57+00:00 0.040600 SELL 12.110856
4 91.455503 2025-09-01 00:03:14+00:00 0.040812 BUY 27.293068
In [33]:
tbls.rename(columns={'Close':'value'}, inplace=True)
tbls.head()
Out[33]:
Date Open High Low value % Change timestamp
0 Sep 03, 2025 23:30 0.039891 0.039893 0.039827 0.039827 -0.16% 2025-09-03 23:30:00+00:00
1 Sep 03, 2025 23:00 0.039893 0.039932 0.039891 0.039891 -0.01% 2025-09-03 23:00:00+00:00
2 Sep 03, 2025 22:30 0.039845 0.039893 0.039845 0.039893 0.12% 2025-09-03 22:30:00+00:00
3 Sep 03, 2025 22:00 0.039820 0.039968 0.039820 0.039845 0.06% 2025-09-03 22:00:00+00:00
4 Sep 03, 2025 21:30 0.039901 0.039901 0.039753 0.039820 -0.20% 2025-09-03 21:30:00+00:00
In [34]:
tbls['timestamp'] = pd.to_datetime(tbls['Date'].values, utc= True)
tbls = tbls[['timestamp', 'value']]
tbls['size'] = 1.1
tbls.head()
Out[34]:
timestamp value size
0 2025-09-03 23:30:00+00:00 0.039827 1.1
1 2025-09-03 23:00:00+00:00 0.039891 1.1
2 2025-09-03 22:30:00+00:00 0.039893 1.1
3 2025-09-03 22:00:00+00:00 0.039845 1.1
4 2025-09-03 21:30:00+00:00 0.039820 1.1
In [35]:
tbls['variable'] = 'huobi_close'
tbls['size_log'] = np.log2(tbls['size'])
In [36]:
df_trds_bids_asks_huobi = pd.concat([df_trds_bids_asks, tbls])

/Users/valdemar/anaconda3/lib/python3.5/site-packages/ipykernel_launcher.py:1: FutureWarning:

Sorting because non-concatenation axis is not aligned. A future version
of pandas will change to not sort by default.

To accept the future behavior, pass 'sort=False'.

To retain the current behavior and silence the warning, pass 'sort=True'.
In [37]:
df_trds_bids_asks_huobi.sort_values(by = 'timestamp', inplace=True)

Analysis and conclusion

In [38]:
trds.groupby('side').count()['size'].plot(kind='bar', title='trades count disbalance by side')
plt.show()

trades disbalance by side

In [39]:
#trades mean size disbalance by side
trds.groupby('side')['size'].mean()
Out[39]:
side
BUY     233.746948
SELL      0.007683
Name: size, dtype: float64
In [40]:
trds[trds['side']=='BUY'].describe()
Out[40]:
price size
count 720.000000 7.200000e+02
mean 0.039853 2.337469e+02
std 0.000641 1.753921e+02
min 0.038920 5.600000e-07
25% 0.039230 9.396000e+01
50% 0.039792 1.881380e+02
75% 0.040169 3.361265e+02
max 0.041031 6.873292e+02
In [41]:
trds[trds['side']=='SELL'].describe()
Out[41]:
price size
count 125.000000 125.000000
mean 0.039702 0.007683
std 0.000553 0.026542
min 0.038800 0.000002
25% 0.039258 0.000323
50% 0.039809 0.001396
75% 0.040079 0.002749
max 0.040961 0.191068

Depth volume ask size stat

In [42]:
#Depth volume ask size stat
df_asks.filter(like='size').sum(axis=1).describe()
Out[42]:
count    188.000000
mean       1.452942
std       14.565507
min        0.318646
25%        0.348571
50%        0.378867
75%        0.399122
max      200.100935
dtype: float64
In [43]:
#Depth volume bids size stat
df_bids.filter(like='size').sum(axis=1).describe()
Out[43]:
count    188.000000
mean       1.531306
std        0.091990
min        1.425538
25%        1.466017
50%        1.507842
75%        1.528932
max        1.798526
dtype: float64
In [44]:
fig = px.scatter(df_melt_bids_asks, title = 'LOB data without trading', x="timestamp", y=[ "value"], color= 'variable')

fig.show()
In [45]:
fig = px.scatter(df_melt_bids_asks.iloc[:np.argmax(df_melt_bids_asks['size'].values)], title = 'LOB data without trading where points are proportional to size(until extremely big ask but small after is the same)', x="timestamp", y=[ "value"], color= 'variable', size='size')

fig.show()
In [46]:
fig = px.scatter(df_trds_bids_asks_huobi[(df_trds_bids_asks_huobi['variable']=='BUY') | (df_trds_bids_asks_huobi['variable']=='SELL') |
                                   (df_trds_bids_asks_huobi['variable']=='bids_price_0') | (df_trds_bids_asks_huobi['variable']=='ask_price_0')].iloc[:], title = 'BID/ASK spread and trades', x="timestamp", y=[ "value"],  color= 'variable')

fig.show()

Conclusion

We can see that buy trading volume is bigger than sell one by several orders of magnitude and volume of LOB(both sides) as well. And obviously there is no significant (depth)volume of ETH in absolute value in the LOB(smaller than 1 ETH for asks side except one outlier and smaller than 1.8ETH for bids side) and LOB levels are extremly regular(especially bids). This can be in the case of something like wash trading activity(without another side or may be no data). I think we don't see those extremely big asks levels(that matches the trades) because the best asks instantaneously match with respective non bona fide buy orders but the date is sparse. But there is one big ask order about 199ETH so it seems that this is that non bona fide ask order because it has the same order of magnitude as median. I think this behavior may pursue several goals. The first and the main(from my point of view) is to emulate big trading activity(volume) of that place for marketing purposes or for rebate, the second is to abuse market participants that market price will go up but anyway there is no such volume in LOB from ask side.(fake arbitrage oportunities).

Auxiliary

It's interesting that there exist theoretic arbitrage oportunities with Huobi for example. But there is now significant volume to realize that as I've already said. This is another reasion that this is manipulative(wash trading) behaviour. We can see in the graph below. I scrap 30min interval OHLC data from Huobi(as they say) for that purpose.

In [47]:
fig = px.scatter(df_trds_bids_asks_huobi[(df_trds_bids_asks_huobi['variable']=='huobi_close') | (df_trds_bids_asks_huobi['variable']=='BUY') | (df_trds_bids_asks_huobi['variable']=='SELL') |
                                   (df_trds_bids_asks_huobi['variable']=='bids_price_0') | (df_trds_bids_asks_huobi['variable']=='ask_price_0')].iloc[:], title = 'BID/ASK spread, trades and Huobi 30min close', x="timestamp", y=[ "value"],  color= 'variable')

fig.show()